Discrete Convexity in Joint Winner Property
|
|
- Christian Poole
- 5 years ago
- Views:
Transcription
1 Discrete Convexity in Joint Winner Property Yuni Iwamasa Kazuo Murota Stanislav Živný January 12, 2018 Abstract In this paper, we reveal a relation between joint winner property (JWP) in the field of valued constraint satisfaction problems (VCSPs) and M -convexity in the field of discrete convex analysis (DCA). We introduce the M -convex completion problem, and show that a function f satisfying the JWP is Z-free if and only if a certain function f associated with f is M -convex completable. This means that if a function is Z-free, then the function can be minimized in polynomial time via M -convex intersection algorithms. Furthermore we propose a new algorithm for Z-free function minimization, which is faster than previous algorithms for some parameter values. Keywords: valued constraint satisfaction problems, discrete convex analysis, M- convexity 1 Introduction A valued constraint satisfaction problem (VCSP) is a general framework for discrete optimization (see [19] for details). Informally, the VCSP framework deals with the minimization problem of a function represented as the sum of small arity functions. It is known that various kinds of combinatorial optimization problems can be formulated in the VCSP framework. In general, the VCSP is NP-hard. An important line of research is to investigate which classes of instances are solvable in polynomial time, and why these classes ensure polynomial time solvability. Cooper Živný [2] showed that if a function represented as the sum of unary or binary functions satisfies the joint winner property (JWP), then the function can be minimized in polynomial time. This gives an example of a class of instances that are solvable in polynomial time. In this paper, we present the reason why JWP ensures polynomial time solvability via discrete convex analysis (DCA) [10], particularly, M -convexity [13]. DCA is a theory of convex functions on discrete structures, and M -convexity is one of the important convexity concepts in DCA. M -convexity appears in many areas such as operations research, economics, and game theory (see e.g., [10, 11, 12]). The results of this paper are summarized as follows: We reveal a relation between JWP and M -convexity. That is, we give a DCA interpretation of polynomial-time solvability of JWP. Department of Mathematical Informatics, Graduate School of Information Science and Technology, University of Tokyo, Tokyo, , Japan. yuni iwamasa@mist.i.u-tokyo.ac.jp Department of Business Administration, Tokyo Metropolitan University, Tokyo, , Japan. murota@tmu.ac.jp Department of Computer Science, University of Oxford, Oxford, OX1 3QD, United Kingdom. standa.zivny@cs.ox.ac.uk 1
2 To describe the connection of JWP and M -convexity, we introduce the M -convex completion problem, and give a characterization of M -convex completability. By utilizing a DCA interpretation of JWP, we propose a new algorithm for Z-free function minimization, which is faster than previous algorithms for some parameter values. This study will hopefully be the first step towards fruitful interactions between VCSPs and DCA. Notations. Let R and R + denote the sets of reals and nonnegative reals, respectively. In this paper, functions can take the infinite value +, where a < +, a + = + for a R, and 0 (+ ) = 0. Let R := R {+ } and R + := R + {+ }. For a function f : {0, 1} n R, the effective domain is denoted as dom f := {x {0, 1} n f(x) < + }. For a positive integer k, we define [k] := {1, 2,..., k}. For x = (x 1, x 2,..., x n ) R n, we define supp + (x) := {i [n] x i > 0}. 2 Preliminaries Joint Winner Property. Let d i 2 be a positive integer and D i := [d i ] for i [r]. We consider a function f : D 1 D 2 D r R + represented as the sum of unary or binary functions as f(x 1, x 2,..., x r ) = c i (x i ) + c ij (x i, x j ), (1) i [r] 1 i<j r where c i : D i R + is a unary function for i [r] and c ij : D i D j R + is a binary function for 1 i < j r. Furthermore we assume c ij = c ji for distinct i, j [r]. A function f of the form (1) is said [2] to satisfy the joint winner property (JWP) if it holds that c ij (a, b) min{c jk (b, c), c ik (a, c)} (2) for all distinct i, j, k [r] and all a D i, b D j, c D k. A function f of the form (1) satisfying the JWP is said to be Z-free if it satisfies that argmin{c ij (a, c), c ij (a, d), c ij (b, c), c ij (b, d)} 2 (3) for any i, j [r] (i j), {a, b} D i (a b), and {c, d} D j (c d). Cooper Živný [2] showed that if f of the form (1) satisfies the JWP, then f can be minimized in polynomial time. In fact, they showed that if f satisfies the JWP, then f can be transformed into a certain Z-free function f in polynomial time such that a minimizer of f is also a minimizer of f. Moreover they showed that a Z-free function can be minimized in polynomial time. JWP appears in many contexts. For example, JWP identifies a tractable class of the MAX- 2SAT problem, which is a well-known NP-hard problem [6]. Indeed, for a 2-CNF formula ψ, we can represent the MAX-2SAT problem for ψ as a Boolean binary {0, 1}-valued VCSP instance. In this binary VCSP instance, JWP is equivalent to the following condition on ψ: if clauses (x 1 x 2 ) and (x 1 x 3 ) are contained in ψ, then so is (x 2 x 3 ). In addition, the AllDifferent constraint [15] and SoftAllDiff constraint [14] can be regarded as special cases of the JWP, and certain scheduling problems introduced in [1, 8] satisfy the JWP. See also Examples 5 8 in [2] for details. 2
3 M -Convexity. A function f : {0, 1} n R is said [10, 11] to be M -convex if for all x, y {0, 1} n and all i supp + (x y) there exists j supp + (y x) {0} such that f(x) + f(y) f(x χ i + χ j ) + f(y + χ i χ j ), (4) where χ i is the ith unit vector and χ 0 is the zero vector. A function f : {0, 1} n R is said [10] to be M 2 -convex if f can be represented as the sum of two M -convex functions. It is well known that M -convex functions can be minimized in polynomial time. Furthermore if we are given two M -convex functions g and h, we can minimize an M 2-convex function f = g + h in polynomial time by solving the so-called M -convex intersection problem. Theorem 1 ([16, Theorem 10][18, Theorem 6.5]; see also [12, Theorem 3.3]). A function f : {0, 1} n R with the zero vector in dom f is M -convex if and only if f satisfies the following two conditions: Condition 1: For all distinct i, j, k [n] and all z {0, 1} n with supp + (z) [n] \ {i, j, k}, it holds that f(z + χ i + χ j ) + f(z + χ k ) min{f(z + χ j + χ k ) + f(z + χ i ), f(z + χ i + χ k ) + f(z + χ j )}. (5) Condition 2: For all distinct i, j [n] and all z {0, 1} n with supp + (z) [n] \ {i, j}, it holds that f(z + χ i + χ j ) + f(z) f(z + χ i ) + f(z + χ j ). We pay special attention to quadratic M -convex functions. Using Theorem 1, we provide a necessary and sufficient condition for the M -convexity of a function f : {0, 1} n R of the form f(x 1, x 2,..., x n ) := h i x i + h ij x i x j ((x 1, x 2,..., x n ) {0, 1} n ), (6) i [n] 1 i<j n where we assume h ij = h ji and h i < + for i, j [n]. Lemma 2. A function f of the form (6) is M -convex if and only if it satisfies the following: h ij min{h ik, h jk } (i, j, k : distinct). h ij 0 (i, j : distinct). In Lemma 2, h ij can take the infinite value +, whereas all h ij s are assumed to be finite in the characterization in [7] and [11]. In particular, we refer to the first condition h ij min{h ik, h jk } (i, j, k : distinct) as the anti-ultrametric property. Note that no conditions are imposed on h i. The proof of Lemma 2 is in Section 5 By Lemma 2, we know that M -convexity of a function of the form (6) depends only on quadratic coefficients (h ij ) i,j [n]. We say that a function f of the form (6) is defined by (h ij ) i,j [n] if the quadratic coefficients of f is equal to (h ij ) i,j [n]. 3 M -Convexity in Joint Winner Property M -Convex Completion Problem. We introduce the M -convex completion problem, and give a characterization of an M -convex completable function on {0, 1} n defined by (h ij ) i,j [n]. The M -convex completion problem is the following: 3
4 Given: (h ij ) i,j [n] such that h ij R or h ij is undefined for every distinct i, j [n]. Question: By assigning appropriate values in R to undefined elements of (h ij ) i,j [n], can we construct an M -convex function f : {0, 1} n R of the form (6)? It should be clear that a defined element can be equal to + and the infinite value (+ ) may be assigned to undefined elements. If there is an appropriate assignment of (h ij ) i,j [n], then (h ij ) i,j [n] is said to be M -convex completable. If h ij < 0 or h ij < min{h jk, h ik } holds for some defined elements h ij, h jk, h ik, then we obviously know that (h ij ) i,j [n] is not M -convex completable. Hence in considering the M -convex completion problem, we assume that h ij 0, (7) h ij min{h jk, h ik } (8) for all defined elements h ij, h jk, h ik. For quadratic coefficients H := (h ij ) i,j [n] containing undefined elements, we define the assignment graph of H as a graph G H = ([n], E H ; w), where E H := {{i, j} i j and h ij is defined} and w : E H R + is defined by w({i, j}) := h ij for {i, j} E H. Then the following theorem holds. Theorem 3. H := (h ij ) i,j [n] is M -convex completable if and only if argmin e C w(e) 2 holds for every chordless cycle C of G H. The proof of Theorem 3 is in Section 5. Remark 4. Farach Kannan Warnow [4] introduced the matrix sandwich problem for ultrametric property, which contains the M -convex completion problem as a special case. They also constructed an O(m + n log n)-time algorithm for the matrix sandwich problem for ultrametric property, where m is the number of defined elements. In our setting, m = O(n 2 ). Hence, by using this algorithm, we can obtain an appropriate M -convex completion in O(n 2 ) time if one exists. An O(n 2 )-time algorithm based on Farach Kannan Warnow s algorithm is the following: Suppose that all h ij are finite (if there exists h ij with h ij = +, then we can redefine the value of h ij as a sufficiently large finite value M). Take any maximum forest F of G H. Let α 1 > α 2 > > α p be the distinct values of defined elements of (h ij ) {i,j} F. For k = 1,..., p 1, let F α k be the subgraph of F induced by the edges with weight at least α k, i.e., F α k := {{i, j} F h ij α k }. Then, for each {i, j} E H with i, j connected in G H, set h ij to α k, where k is the minimum number such that i, j is connected in F α k. For each {i, j} E H with i, j disconnected in G H, set h ij to α p. In this paper, we present a graphic characterization of M -convex completability. With this characterization, we provide a DCA interpretation of polynomial-time solvability of JWP. Transformation into a Function over {0, 1}. To connect JWP and M -convexity, we introduce a transformation of a function f : D 1 D 2 D r R into a function ˆf : {0, 1} U R, where U is the set of all assignments to variables, that is, U := {(1, 1), (1, 2),..., (1, d 1 ), (2, 1), (2, 2),..., (2, d 2 ),..., (r, 1), (r, 2),..., (r, d r )}. We consider the following correspondence between x = (x 1, x 2,..., x r ) D 1 D 2 D r and ˆx = (ˆx (1,1),..., ˆx (1,d1 ), ˆx (2,1),..., ˆx (2,d2 ),..., ˆx (r,1),..., ˆx (r,dr)) {0, 1} U : (1,x 1 ) (2,x 2 ) (r,x r) (x 1, x 2,..., x r ) (0,..., 0, ˇ1, 0,..., 0, 0,..., 0, ˇ1, 0,..., 0,..., 0,..., 0, ˇ1, 0,..., 0). }{{}}{{}}{{} d 1 d 2 d r (9) 4
5 That is, ˆx (i,a) = 1 means that we assign a to x i, and ˆx (i,a) = 0 means that we do not. In view of (9), define a function ˆf by { f(x) if there exists x satisfying (9), ˆf(ˆx) := (ˆx {0, 1} U ). + otherwise Note that minimizing f is equivalent to minimizing ˆf. Now we consider the transformation of f of the form (1) into ˆf, where f is given in terms of c i for i [r] and c ij for i, j [r]. We define f : {0, 1} U R + by f(ˆx) := c i (a)ˆx (i,a) + h (i,a),(j,b)ˆx (i,a)ˆx (j,b) (ˆx {0, 1} U ), (10) (i,a) U (i,a),(j,b) U, (i,a) (j,b) where h (i,a),(j,b) := { c ij (a, b) if i j, undefined if i = j. (11) We also define δ U : {0, 1} U R by { 0 if there exists x satisfying (9), δ U (ˆx) := + otherwise (ˆx {0, 1} U ), which is the indicator function for the feasible assignments. Then we have ˆf(ˆx) = f(ˆx) + δ U (ˆx) (ˆx {0, 1} U ), where arbitrary values in R may be assigned to the undefined elements h (i,a),(i,b) in f without affecting the value of ˆf. Indeed, if ˆx dom δu, then ˆx (i,a)ˆx (i,b) = 0 for all i [r] and all distinct a, b D i. Hence h (i,a),(i,b)ˆx (i,a)ˆx (i,b) = 0 holds for each undefined element h (i,a),(i,b) by the definition (11). In particular, the set of minimizers of both ˆf(ˆx) and f(ˆx) + δ U (ˆx) are the same. It is clear that δ U is M -convex (dom δ U is the base family of a partition matroid, which is a direct sum of matroids of rank 1). Hence if (h (i,a),(j,b) ) (i,a),(j,b) U has an M -convex completion ( h (i,a),(j,b) ) (i,a),(j,b) U, then f defined by ( h (i,a),(j,b) ) (i,a),(j,b) U is M -convex and ˆf = f + δ U is M 2 -convex. This means that ˆf can be minimized in polynomial time. We need the values of ( h (i,a),(j,b) ) (i,a),(j,b) U in a minimization algorithm of M 2-convex functions. A function of the form (1) satisfies the JWP if and only if h (i,a),(j,b) min{h (j,b),(k,c), h (i,a),(k,c) } holds for defined elements h (i,a),(j,b), h (j,b),(k,c), h (i,a),(k,c) given in (10). Hence (h (i,a),(j,b) ) (i,a),(j,b) U satisfies the assumptions (7) and (8) for the M -convex completion problem. Theorem 3 implies the following theorem (the proof is in Section 5). Theorem 5. For a function f of the form (1), let (h (i,a),(j,b) ) (i,a),(j,b) U be defined by (11). Then (h (i,a),(j,b) ) (i,a),(j,b) U is M -convex completable if and only if f (has the JWP and) is Z-free. 4 Algorithm By using a general algorithm for the M -convex intersection (minimization of M 2-convex functions), we can minimize Z-free functions of the form (1) in polynomial time. Suppose that we are given c i : D i R + for i [r], c ij : D i D j R + for 1 i < j r, and a Z-free function 5
6 f defined as (1). We can minimize f by minimizing ˆf = f + δ U with an M -convex intersection algorithm. Here we take advantage of the fact that all the vectors in dom δ U have a constant component sum, i.e., (i,a) U ˆx (i,a) = r for all ˆx dom δ U. This implies that δ U is an M-convex function [10] and we can use an M-convex intersection algorithm. An M-convex intersection algorithm is easier to describe than an M -convex intersection algorithm, though the time complexity is the same. Therefore we devise a minimization algorithm for Z-free functions via an M-convex intersection algorithm. Since the functions are defined on {0, 1} n, the proposed algorithm is actually a variant of valuated matroid intersection algorithms [9]. Specifically, let f r denote the restriction of f to the hyperplane containing dom δ U, i.e., f(ˆx) if ˆx (i,a) = r, f r (ˆx) := (i,a) U + otherwise. Then minimizing f + δ U is equivalent to minimizing f r + δ U, where f r and δ U are M-convex functions. The proposed algorithm consists of three steps. Step 1: On the basis of Theorem 5, we construct an M -convex function f : {0, 1} U R in (10) through an M -convex completion of (h (i,a),(j,b) ) (i,a),(j,b) U in (11). Step 2: We find a minimizer of f r, to be used as an initial solution in Step 3. Step 3: We find a minimizer of f r +δ U by the successive shortest path algorithm with potentials for the M-convex intersection [10] (see also [9, Section 5.2]). In Step 3 of the algorithm, we use the auxiliary graph Gˆx,ŷ = (V, Eˆx,ŷ ) defined for ˆx dom f r and ŷ dom δ U by V := {s, t} U, (12) Eˆx := {((i, a), (j, b)) (i, a), (j, b) U, ˆx + χ (j,b) χ (i,a) dom f r }, (13) Eŷ := {((i, a), (j, b)) (i, a), (j, b) U, ŷ + χ (i,a) χ (j,b) dom δ U }, (14) E + := {(s, (i, a)) (i, a) supp + (ˆx ŷ)}, (15) E := {((j, b), t) (j, b) supp + (ŷ ˆx)}, (16) Eˆx,ŷ := Eˆx Eŷ E + E (17) with the arc length function l = lˆx,ŷ : Eˆx,ŷ R given by { f r (ˆx + χ v χ u ) f r (ˆx) if (u, v) Eˆx, l(u, v) := 0 otherwise. (18) Note that, by the definition of δ U, we can also describe Eŷ as Eŷ = {((i, a), (i, b)) i [r], a, b D i, (i, a) supp + (ŷ), (i, b) supp + (ŷ)}. Algorithm for Z-free function minimization: Step 1: Find an M -convex completion ( h (i,a),(j,b) ) (i,a),(j,b) U of (h (i,a),(j,b) ) (i,a),(j,b) U, and define f : {0, 1} U R by f(ˆx) = c i (a)ˆx (i,a) + h (i,a),(j,b)ˆx (i,a)ˆx (j,b) (ˆx {0, 1} U ). (i,a) U (i,a),(j,b) U, (i,a) (j,b) 6
7 Step 2: Let ˆx {0, 1} U be the zero vector. While (i,a) U ˆx (i,a) < r, do the following: Step 2-1: Obtain (i, a) argmin{f(ˆx + χ (i,a) ) (i, a) U \ supp + (ˆx )}. Step 2-2: ˆx ˆx + χ (i,a). Step 3: Let p : V R be a potential defined by p(v) := 0 for v {s, t} U. ŷ dom δ U. While ˆx ŷ, do the following: Take any Step 3-1: Make the auxiliary graph Gˆx,ŷ. Define the modified arc length l p : Eˆx,ŷ R by l p (u, v) := l(u, v) + p(u) p(v) for (u, v) Eˆx,ŷ. Step 3-2: For each v V, compute the length p(v) of an s-v shortest path in Gˆx,ŷ with respect to the modified arc length l p. Let P be an s-t shortest path having the smallest number of arcs in Gˆx,ŷ with respect to the modified arc length l p. Step 3-3: For (i, a) U, ˆx (i,a) 1 if ((i, a), (j, b)) P Eˆx, ˆx (i,a) ˆx (i,a) + 1 if ((j, b), (i, a)) P Eˆx, ˆx (i,a) otherwise, ŷ(i,a) + 1 if ((i, a), (j, b)) P E ŷ, ŷ(i,a) ŷ(i,a) 1 if ((j, b), (i, a)) P E ŷ, otherwise. ŷ (i,a) For v V, p(v) p(v) + p(v). At the end of Step 2, we obtain a minimizer of f r. The validity of Step 2 is given in [13, Theorem 3.2]. The time complexity of this algorithm is as follows, where n := U = i [r] d i (the proof is in Section 5). Theorem 6. The proposed algorithm runs in O(nr 3 + nr log n + n 2 ) time. By improving the algorithm of running time O(n 3 ) given in [2], Cooper Živný [3] gave an O(n 2 log n log r)-time algorithm for minimizing Z-free functions of the form (1). Our proposed algorithm is faster than Cooper Živný s for some r (e.g., r = O(n1/3 )). Remark 7. In the VCSP framework, we assume that the function f of the form (1) is explicitly given. This means that the input size is proportional to d i + log c i (d) + log c ij (d, e), i [r] i [r] d D i 1 i<j r e D j and then the running time in Theorem 6 is strongly polynomial in the input size. On the other hand, if we assume that f is given by the value oracles for the functions c i and c ij, the input size of f is proportional to r + log d i + log c i (d) + log c ij (d, e). i [r] i [r] d D i 1 i<j r e D j In this case, the running time in Theorem 6 is pseudo-polynomial in the input size. d D i d D i 7
8 5 Proofs In this section, we give the proofs of Lemma 2, Theorem 3, Theorem 5, and Theorem 6. Proof of Lemma 2. (only-if part). Suppose that there exist distinct i, j, k [n] such that h ij < min{h jk, h ik }. Note that h ij < + holds. Then f(χ i + χ j ) + f(χ k ) = h i + h j + h k + h ij < h i + h j + h k + h jk = f(χ j + χ k ) + f(χ i ), f(χ i + χ j ) + f(χ k ) = h i + h j + h k + h ij < h i + h j + h k + h ik = f(χ i + χ k ) + f(χ j ) hold since h i, h j, h k, h ij < +. By Condition 1 of Theorem 1, f is not M -convex. Suppose that there exist distinct i, j [n] such that h ij < 0(< + ). Then f(χ i + χ j ) + f(χ 0 ) = h i + h j + h ij < h i + h j = f(χ i ) + f(χ j ) holds since h i, h j < +. By Condition 2 of Theorem 1, f is not M -convex. (if part). Take arbitrary distinct i, j, k [n] and z {0, 1} n with supp + (z) [n] \ {i, j, k}. If f(z + χ i + χ j ) = + or f(z + χ k ) = + holds, then Condition 1 of Theorem 1 obviously holds. We assume f(z + χ i + χ j ) < + and f(z + χ k ) < +. It holds that f(z + χ i + χ j ) = f(z) + h i + h j + h ip + h jp + h ij, (19) f(z + χ k ) = f(z) + h k + h kp. (20) Note that all terms appearing in (19) and (20) have finite values since f(z + χ i + χ j ) < + and f(z + χ k ) < + hold. Then we have Also we have f(z + χ i + χ j ) + f(z + χ k ) f(z + χ j + χ k ) + f(z + χ i ) 2f(z) + h i + h j + h k + h ip + h jp + 2f(z) + h j + h k + h i + h ij h jk. h jp + h kp + f(z + χ i + χ j ) + f(z + χ k ) f(z + χ i + χ k ) + f(z + χ j ) h ij h ik. By the assumption, it holds that h ij min{h jk, h ik }. Hence we obtain h kp + h ij h ip + h jk f(z + χ i + χ j ) + f(z + χ k ) min{f(z + χ j + χ k ) + f(z + χ i ), f(z + χ i + χ k ) + f(z + χ j )}. By the assumption of h ij 0, we also obtain for all distinct i, j [n]. f(z + χ i + χ j ) + f(z) f(z + χ i ) + f(z + χ j ) 8
9 Proof of Theorem 3. First we give a graphical interpretation for the anti-ultrametric property. For H := (h ij ) i,j [n] and α R, let us define EH α and V H α by Let G α H := (V H α, Eα H ). Then the following lemma holds: E α H := {{i, j} E H h ij α}, (21) V α H := {i e E α H such that i e}. (22) Lemma 8. H := (h ij ) i,j [n] satisfies the anti-ultrametric property if and only if each connected component of G α H is a complete graph for every α R. Proof. (only-if part). We show the contraposition. Suppose that for some α R there exists a non-complete graph among the connected components of G α H. Then there exist distinct i, j, k [n] with {i, j}, {j, k} EH α {i, k}. By the definition of Eα H, it holds that min{h ij, h jk } α > h ik. This means that {h ij, h jk, h ik } does not satisfy the anti-ultrametric property. (if part). Suppose that each connected component of G α H is a complete graph for all α R. To show the anti-ultrametric property of (h ij ) i,j [n], it suffices to prove h jk = h ik for all distinct i, j, k satisfying h ij > h jk. If h ik h ij, then there exists a non-complete graph among the connected components of G α H for α = h ij, which is a contradiction. If h ij > h ik > h jk, then there exists a non-complete graph among the connected components of G α H for α = h ik, which is a contradiction. If h jk > h ik, then there exists a non-complete graph among the connected components of G α H for α = h jk, which is a contradiction. Therefore we must have h jk = h ik. We are now ready to prove Theorem 3. Proof of Theorem 3. (only-if part). Suppose to the contrary that H := (h ij ) i,j [n] is M -convex completable and that there exists a chordless cycle C of G H with argmin e C w(e) = 1. Let C = {{i 1, i 2 }, {i 2, i 3 },..., {i m, i 1 }}, and consider the corresponding entries {h i1 i 2, h i2 i 3,..., h imi1 } of H. Note that h ipiq is undefined for p, q [m] with p q = 1 mod m. We may assume α := h i1 i 2 = min{h i1 i 2, h i2 i 3,..., h imi1 }. By the assumption of argmin e C w(e) = 1, we have min{h i2 i 3,..., h imi1 } > α. Since {h i1 i 2, h i2 i 3, h i1 i 3 } should satisfy the anti-ultrametric property, we have to assign α to h i1 i 3 to obtain an M -convex completion. Since {h i1 i 3, h i3 i 4, h i1 i 4 } should satisfy the anti-ultrametric property, we have to assign α to h i1 i 3 to obtain an M - convex completion. By repeating this procedure, we arrive at h i1 i m 1 = α. This is a contradiction, since h i1 i m 1 < min{h im 1 i m, h i1 i m } and hence the anti-ultrametric property fails for {h i1 i m 1, h im 1 i m, h imi1 }. (if part). For α R +, define SH α by S α H := {{i, j} E H i, j V α H, i and j are connected in G α H}. Let G α H := (V H α, Eα H Sα H ). Recall that Eα H and V H α are defined in (21) and (22). First we show that if each connected component of G α H is a complete graph for every α R +, then H is M -convex completable. Let α 1 > α 2 > > α p be the distinct values of defined elements of (h ij ) i,j [n] (α 1 can be the infinite value). We assign α 1 to each undefined element h ij such that {i, j} S α 1 H, α k to each h ij such that {i, j} S α k H \ Sα k 1 H for k = 2,..., p 1, and α p to each h ij such that {i, j} S α p 1 H. Then we obtain a certain completion H := ( h ij ) i,j [n] of (h ij ) i,j [n]. It is clear that each connected component of G α H is a complete graph for every α R. By Lemma 8, H satisfies the anti-ultrametric property. This means that H is M -convex completable. Next we show that if argmin e C w(e) 2 holds for every chordless cycle C of G H, then each connected component of G α H is a complete graph for every α R +. Take arbitrary α R + 9
10 and i and j which are connected in G α H (Note that vertex sets of connected components of G α H are the same as those of G α H ). It suffices to prove that {i, j} Eα H or {i, j} Sα H holds. Suppose to the contrary that there exist i and j such that {i, j} EH α and {i, j} Sα H hold. Let I be the set of such {i, j}. Let {i 0, j 0 } I be a pair of vertices such that the number of edges of a shortest i 0 -j 0 path on G α H is minimum in I. Since i 0 and j 0 are connected in G α H and {i 0, j 0 } SH α, we have {i 0, j 0 } E H. Moreover since {i 0, j 0 } EH α, h i 0 j 0 < α holds. Take a i 0 -j 0 shortest path P 0. Then P 0 {i 0, j 0 } is a chordless cycle of G H. Indeed, if P 0 {i 0, j 0 } has a chord in G H, there exist i and j satisfying {i, j } {i 0, j 0 } in P 0 {i 0, j 0 } such that EH α {i, j } E H by the minimality of I. Then {i, j } I and the number of edges of a shortest i -j path is smaller that those of P 0. However this is a contradiction to the minimality of {i 0, j 0 }. Hence P 0 {i 0, j 0 } is a chordless cycle of G H. It holds that h ij α for {i, j} P 0 and h i0 j 0 < α. Therefore we obtain argmin e P0 {i 0,j 0 } w(e) = 1. This contracts the assumption of argmin e C w(e) 2. Hence we have {i, j} EH α or {i, j} Sα H. Proof of Theorem 5. Let H := (h (i,a),(j,b) ) (i,a),(j,b) U, where the entries h (i,a),(j,b) with i = j are undefined. Recall that G H = (U, E H ; w) is the assignment graph of H. By the definition of E H and h (i,a),(j,b) in (11), we have E H = {{(i, a), (j, b)} i j, a D i, b D j }. By Theorem 3, H is M -convex completable if and only if every chordless cycle C satisfies the condition argmin e C w(e) 2. First we show that chordless cycles in G H have length 3 or 4. Take any chordless cycle C = {{(i 1, a 1 ), (i 2, a 2 )}, {(i 2, a 2 ), (i 3, a 3 )},..., {(i k, a k ), (i 1, a 1 )}} of G H. Since C is chordless, we have i 1 = i p for 3 p k 1 and i 2 = i q for 4 q k. This implies k 4, since otherwise we obtain i 1 = i 4 = i 2, contradicting the existence of an edge between (i 1, a 1 ) and (i 2, a 2 ). For a (chordless) cycle of length 3, say, C = {{(i 1, a 1 ), (i 2, a 2 )}, {(i 2, a 2 ), (i 3, a 3 )}, {(i 3, a 3 ), (i 1, a 1 )}} with i 1 i 2 i 3 i 1, the condition argmin e C w(e) 2 is equivalent to (2) for JWP. For a chordless cycle of length 4, say, C = {{(i 1, a 1 ), (i 2, a 2 )}, {(i 2, a 2 ), (i 3, a 3 )}, {(i 3, a 3 ), (i 4, a 4 )}, {(i 4, a 4 ), (i 1, a 1 )}} we have i 1 i 2, i 3 i 4, i 1 = i 3, i 2 = i 4, a 1 a 3, a 2 a 4, and then the condition argmin e C w(e) 2 is equivalent to (3) for Z-freeness. Proof of Theorem 6. We investigate each step in turn. (Step 1). Since the number of defined elements of (h (i,a),(j,b) ) (i,a),(j,b) U is O(n 2 r i=1 d2 i ) = O(n 2 ), we can find an M -convex completion in O(n 2 + n log n) time (recall Remark 4). (Step 2). If we have the value of f(ˆx ), we can compute the value of f(ˆx + χ (i,a) ) in O(r) time since f(ˆx + χ (i,a) ) = f(ˆx ) + c i (a) + h (j,b) supp + (ˆx ) (i,a),(j,b). Hence the time complexity of Step 3 is O(nr 2 ) time. (Step 3). Recall the definition of Gˆx,ŷ in (12) (18). We have Eˆx = O(r(n r)) = O(nr), Eŷ = O(n), E + = O(r), and E = O(r). Hence Eˆx,ŷ = O(nr). Furthermore we need to compute l only on Eˆx, since l is equal to zero on other arcs. If we have the value of f(ˆx ) at hand, we can compute the value of f(ˆx + χ (j,b) χ (i,a) ) in O(r) time since f(ˆx + χ (j,b) χ (i,a) ) = f(ˆx ) c i (a) + (k,c) supp + (ˆx ) h (i,a),(k,c) + c j (b) + (k,c) supp + (ˆx χ (i,a) ) h (j,b),(k,c). Therefore we can construct the auxiliary graph Gˆx,ŷ in O(nr2 ) time. The modified arc length l p is nonnegative [9, Section 5.2]. Hence we can compute p(v) for v V and a shortest path P in Step 3-2 in O(nr + n log n) time by using Dijkstra s algorithm with Fibonacci heaps [5] (see also [17, Section 7.4]). We can update ˆx, ŷ, and p in Step 3-3 in 10
11 O(nr) time. By one iteration of Step 3, the value of ˆx ŷ 1 is decreased by two. Hence the number of iterations of Step 3 is bounded by O(r). Therefore the time complexity of Step 3 is O(nr 3 + nr log n). By the above argument, we see that the proposed algorithm runs in O(nr 3 + nr log n + n 2 ) time. Acknowledgments We thank Kazutoshi Ando and Takanori Maehara for information on the paper [4] in Remark 4. We also thank the referees for helpful comments. This research was initiated at the Trimester Program Combinatorial Optimization at Hausdorff Institute of Mathematics, The first author s research was supported by JSPS Research Fellowship for Young Scientists. The second author s research was supported by The Mitsubishi Foundation, CREST, JST, and JSPS KAK- ENHI Grant Number The last author s research was supported by a Royal Society University Research Fellowship. This project has received funding from the European Research Council (ERC) under the European Union s Horizon 2020 research and innovation programme (grant agreement No ). The paper reflects only the authors views and not the views of the ERC or the European Commission. The European Union is not liable for any use that may be made of the information contained therein. References [1] J. L. Bruno, E. G. Coffman Jr., and R. Sethi. Scheduling independent tasks to reduce mean finishing time. Communications of the ACM, 17(7): , [2] M. C. Cooper and S. Živný. Hybrid tractability of valued constraint problems. Artificial Intelligence, 175: , [3] M. C. Cooper and S. Živný. Tractable triangles and cross-free convexity in discrete optimisation. Journal of Artificial Intelligence Research, 44: , [4] M. Farach, S. Kannan, and T. Warnow. A robust model for finding optimal evolutionary trees. Algorithmica, 13: , [5] M. L. Fredman and R. E. Tarjan. Fibonacci heaps and their uses in improved network optimization algorithms. Journal of the Association for Computing Machinery, 34: , [6] M. R. Garey, D. S. Johnson, and L. Stockmeyer. Some simplified NP-complete graph problems. Theoretical Computer Science, 1: , [7] H. Hirai and K. Murota. M-convex functions and tree metrics. Japan Journal of Industrial and Applied Mathematics, 21: , [8] W. A. Horn. Minimizing average flow time with parallel machines. Operations Research, 21(3): , [9] K. Murota. Matrices and Matroids for Systems Analysis. Springer, Heidelberg, [10] K. Murota. Discrete Convex Analysis. SIAM, Philadelphia,
12 [11] K. Murota. Recent developments in discrete convex analysis. In W. Cook, L. Lovász, and J. Vygen, editors, Research Trends in Combinatorial Optimization, chapter 11, pages Springer, Heidelberg, [12] K. Murota. Discrete convex analysis: A tool for economics and game theory. Journal of Mechanism and Institution Design, 1(1): , [13] K. Murota and A. Shioura. M-convex function on generalized polymatroid. Mathematics of Operations Research, 24(1):95 105, [14] T. Petit, J. C. Régin, and C. Bessière. Specific filtering algorithms for over-constrained problems. In Proceedings of the 7th International Conference on Principles and Practice of Constraint Programming (CP 01), pages , Heidelberg, Springer. [15] J. C. Régin. A filtering algorithm for constraints of difference in CSPs. In Proceedings of the 12th National Conference on Artificial Intelligence (AAAI 94), volume 1, pages , [16] H. Reijnierse, A. van Gellekom, and J. A. M. Potters. Verifying gross substitutability. Economic Theory, 20: , [17] A. Schrijver. Combinatorial Optimization: Polyhedra and Efficiency. Springer, Heidelberg, [18] A. Shioura and A. Tamura. Gross substitutes condition and discrete concavity for multi-unit valuations: a survey. Journal of the Operations Research Society of Japan, 58(1):61 103, [19] S. Živný. The Complexity of Valued Constraint Satisfaction Problems. Springer, Heidelberg,
Beyond JWP: A Tractable Class of Binary VCSPs via M-Convex Intersection
Beyond JWP: A Tractable Class of Binary VCSPs via M-Convex Intersection Hiroshi Hirai Department of Mathematical Informatics, University of Tokyo, Japan hirai@mist.i.u-tokyo.ac.jp Yuni Iwamasa Department
More informationHybrid tractability of valued constraint problems
Hybrid tractability of valued constraint problems Martin C. Cooper a, Stanislav Živný,b,c a IRIT, University of Toulouse III, 31062 Toulouse, France b University College, University of Oxford, OX1 4BH
More informationMinimization and Maximization Algorithms in Discrete Convex Analysis
Modern Aspects of Submodularity, Georgia Tech, March 19, 2012 Minimization and Maximization Algorithms in Discrete Convex Analysis Kazuo Murota (U. Tokyo) 120319atlanta2Algorithm 1 Contents B1. Submodularity
More informationON DISCRETE HESSIAN MATRIX AND CONVEX EXTENSIBILITY
Journal of the Operations Research Society of Japan Vol. 55, No. 1, March 2012, pp. 48 62 c The Operations Research Society of Japan ON DISCRETE HESSIAN MATRIX AND CONVEX EXTENSIBILITY Satoko Moriguchi
More informationMATHEMATICAL ENGINEERING TECHNICAL REPORTS. Dijkstra s Algorithm and L-concave Function Maximization
MATHEMATICAL ENGINEERING TECHNICAL REPORTS Dijkstra s Algorithm and L-concave Function Maximization Kazuo MUROTA and Akiyoshi SHIOURA METR 2012 05 March 2012; revised May 2012 DEPARTMENT OF MATHEMATICAL
More informationDiscrete Convex Analysis
RIMS, August 30, 2012 Introduction to Discrete Convex Analysis Kazuo Murota (Univ. Tokyo) 120830rims 1 Convexity Paradigm NOT Linear vs N o n l i n e a r BUT C o n v e x vs Nonconvex Convexity in discrete
More informationThe Complexity of Maximum. Matroid-Greedoid Intersection and. Weighted Greedoid Maximization
Department of Computer Science Series of Publications C Report C-2004-2 The Complexity of Maximum Matroid-Greedoid Intersection and Weighted Greedoid Maximization Taneli Mielikäinen Esko Ukkonen University
More informationOn Steepest Descent Algorithms for Discrete Convex Functions
On Steepest Descent Algorithms for Discrete Convex Functions Kazuo Murota Graduate School of Information Science and Technology, University of Tokyo, and PRESTO, JST E-mail: murota@mist.i.u-tokyo.ac.jp
More informationReduction of Ultrametric Minimum Cost Spanning Tree Games to Cost Allocation Games on Rooted Trees. Kazutoshi ANDO and Shinji KATO.
RIMS-1674 Reduction of Ultrametric Minimum Cost Spanning Tree Games to Cost Allocation Games on Rooted Trees By Kazutoshi ANDO and Shinji KATO June 2009 RESEARCH INSTITUTE FOR MATHEMATICAL SCIENCES KYOTO
More informationinforms DOI /moor.xxxx.xxxx
MATHEMATICS OF OPERATIONS RESEARCH Vol. 00, No. 0, Xxxxxx 20xx, pp. xxx xxx ISSN 0364-765X EISSN 1526-5471 xx 0000 0xxx informs DOI 10.1287/moor.xxxx.xxxx c 20xx INFORMS Polynomial-Time Approximation Schemes
More informationA necessary and sufficient condition for the existence of a spanning tree with specified vertices having large degrees
A necessary and sufficient condition for the existence of a spanning tree with specified vertices having large degrees Yoshimi Egawa Department of Mathematical Information Science, Tokyo University of
More informationMATHEMATICAL ENGINEERING TECHNICAL REPORTS. On the Pipage Rounding Algorithm for Submodular Function Maximization A View from Discrete Convex Analysis
MATHEMATICAL ENGINEERING TECHNICAL REPORTS On the Pipage Rounding Algorithm for Submodular Function Maximization A View from Discrete Convex Analysis Akiyoshi SHIOURA (Communicated by Kazuo MUROTA) METR
More informationM-convex Functions on Jump Systems A Survey
Discrete Convexity and Optimization, RIMS, October 16, 2012 M-convex Functions on Jump Systems A Survey Kazuo Murota (U. Tokyo) 121016rimsJUMP 1 Discrete Convex Analysis Convexity Paradigm in Discrete
More informationMATHEMATICAL ENGINEERING TECHNICAL REPORTS. Induction of M-convex Functions by Linking Systems
MATHEMATICAL ENGINEERING TECHNICAL REPORTS Induction of M-convex Functions by Linking Systems Yusuke KOBAYASHI and Kazuo MUROTA METR 2006 43 July 2006 DEPARTMENT OF MATHEMATICAL INFORMATICS GRADUATE SCHOOL
More informationGross Substitutes Tutorial
Gross Substitutes Tutorial Part I: Combinatorial structure and algorithms (Renato Paes Leme, Google) Part II: Economics and the boundaries of substitutability (Inbal Talgam-Cohen, Hebrew University) Three
More informationA Min-Max Theorem for k-submodular Functions and Extreme Points of the Associated Polyhedra. Satoru FUJISHIGE and Shin-ichi TANIGAWA.
RIMS-1787 A Min-Max Theorem for k-submodular Functions and Extreme Points of the Associated Polyhedra By Satoru FUJISHIGE and Shin-ichi TANIGAWA September 2013 RESEARCH INSTITUTE FOR MATHEMATICAL SCIENCES
More informationDiscrete DC Programming by Discrete Convex Analysis
Combinatorial Optimization (Oberwolfach, November 9 15, 2014) Discrete DC Programming by Discrete Convex Analysis Use of Conjugacy Kazuo Murota (U. Tokyo) with Takanori Maehara (NII, JST) 141111DCprogOberwolfLong
More informationMATHEMATICAL ENGINEERING TECHNICAL REPORTS. Discrete Hessian Matrix for L-convex Functions
MATHEMATICAL ENGINEERING TECHNICAL REPORTS Discrete Hessian Matrix for L-convex Functions Satoko MORIGUCHI and Kazuo MUROTA METR 2004 30 June 2004 DEPARTMENT OF MATHEMATICAL INFORMATICS GRADUATE SCHOOL
More informationMATHEMATICAL ENGINEERING TECHNICAL REPORTS
MATHEMATICAL ENGINEERING TECHNICAL REPORTS Combinatorial Relaxation Algorithm for the Entire Sequence of the Maximum Degree of Minors in Mixed Polynomial Matrices Shun SATO (Communicated by Taayasu MATSUO)
More informationScheduling Parallel Jobs with Linear Speedup
Scheduling Parallel Jobs with Linear Speedup Alexander Grigoriev and Marc Uetz Maastricht University, Quantitative Economics, P.O.Box 616, 6200 MD Maastricht, The Netherlands. Email: {a.grigoriev, m.uetz}@ke.unimaas.nl
More informationComputing DM-decomposition of a partitioned matrix with rank-1 blocks
Computing DM-decomposition of a partitioned matrix with rank-1 blocks arxiv:1609.01934v2 [math.co] 17 Feb 2018 Hiroshi HIRAI Department of Mathematical Informatics, Graduate School of Information Science
More informationA Note on Discrete Convexity and Local Optimality
A Note on Discrete Convexity and Local Optimality Takashi Ui Faculty of Economics Yokohama National University oui@ynu.ac.jp May 2005 Abstract One of the most important properties of a convex function
More informationA combinatorial algorithm minimizing submodular functions in strongly polynomial time
A combinatorial algorithm minimizing submodular functions in strongly polynomial time Alexander Schrijver 1 Abstract We give a strongly polynomial-time algorithm minimizing a submodular function f given
More informationMAKING BIPARTITE GRAPHS DM-IRREDUCIBLE
SIAM J. DISCRETE MATH. Vol. 32, No. 1, pp. 560 590 c 2018 Society for Industrial and Applied Mathematics MAKING BIPARTITE GRAPHS DM-IRREDUCIBLE KRISTÓF BÉRCZI, SATORU IWATA, JUN KATO, AND YUTARO YAMAGUCHI
More informationarxiv: v3 [math.oc] 28 May 2017
On a general framework for network representability in discrete optimization Yuni Iwamasa September 3, 2018 arxiv:1609.03137v3 [math.oc] 28 May 2017 Abstract In discrete optimization, representing an objective
More informationDetecting Backdoor Sets with Respect to Horn and Binary Clauses
Detecting Backdoor Sets with Respect to Horn and Binary Clauses Naomi Nishimura 1,, Prabhakar Ragde 1,, and Stefan Szeider 2, 1 School of Computer Science, University of Waterloo, Waterloo, Ontario, N2L
More informationLecture 10 (Submodular function)
Discrete Methods in Informatics January 16, 2006 Lecture 10 (Submodular function) Lecturer: Satoru Iwata Scribe: Masaru Iwasa and Yukie Nagai Submodular functions are the functions that frequently appear
More informationExact Bounds for Steepest Descent Algorithms of L-convex Function Minimization
Exact Bounds for Steepest Descent Algorithms of L-convex Function Minimization Kazuo Murota Graduate School of Information Science and Technology, University of Tokyo, Tokyo 113-8656, Japan Akiyoshi Shioura
More informationMATHEMATICAL ENGINEERING TECHNICAL REPORTS. Deductive Inference for the Interiors and Exteriors of Horn Theories
MATHEMATICAL ENGINEERING TECHNICAL REPORTS Deductive Inference for the Interiors and Exteriors of Horn Theories Kazuhisa MAKINO and Hirotaka ONO METR 2007 06 February 2007 DEPARTMENT OF MATHEMATICAL INFORMATICS
More informationComplexity Theory VU , SS The Polynomial Hierarchy. Reinhard Pichler
Complexity Theory Complexity Theory VU 181.142, SS 2018 6. The Polynomial Hierarchy Reinhard Pichler Institut für Informationssysteme Arbeitsbereich DBAI Technische Universität Wien 15 May, 2018 Reinhard
More informationOutline. Complexity Theory EXACT TSP. The Class DP. Definition. Problem EXACT TSP. Complexity of EXACT TSP. Proposition VU 181.
Complexity Theory Complexity Theory Outline Complexity Theory VU 181.142, SS 2018 6. The Polynomial Hierarchy Reinhard Pichler Institut für Informationssysteme Arbeitsbereich DBAI Technische Universität
More informationRevisiting the Limits of MAP Inference by MWSS on Perfect Graphs
Revisiting the Limits of MAP Inference by MWSS on Perfect Graphs Adrian Weller University of Cambridge CP 2015 Cork, Ireland Slides and full paper at http://mlg.eng.cam.ac.uk/adrian/ 1 / 21 Motivation:
More informationReachability-based matroid-restricted packing of arborescences
Egerváry Research Group on Combinatorial Optimization Technical reports TR-2016-19. Published by the Egerváry Research Group, Pázmány P. sétány 1/C, H 1117, Budapest, Hungary. Web site: www.cs.elte.hu/egres.
More informationThe Maximum Flow Problem with Disjunctive Constraints
The Maximum Flow Problem with Disjunctive Constraints Ulrich Pferschy Joachim Schauer Abstract We study the maximum flow problem subject to binary disjunctive constraints in a directed graph: A negative
More informationL-convexity on graph structures
L-convexity on graph structures Hiroshi HIRAI Department of Mathematical Informatics, Graduate School of Information Science and Technology, University of Tokyo, Tokyo, 113-8656, Japan. hirai@mist.i.u-tokyo.ac.jp
More informationMulticommodity Flows and Column Generation
Lecture Notes Multicommodity Flows and Column Generation Marc Pfetsch Zuse Institute Berlin pfetsch@zib.de last change: 2/8/2006 Technische Universität Berlin Fakultät II, Institut für Mathematik WS 2006/07
More informationThe cocycle lattice of binary matroids
Published in: Europ. J. Comb. 14 (1993), 241 250. The cocycle lattice of binary matroids László Lovász Eötvös University, Budapest, Hungary, H-1088 Princeton University, Princeton, NJ 08544 Ákos Seress*
More informationComputational Complexity
Computational Complexity Algorithm performance and difficulty of problems So far we have seen problems admitting fast algorithms flow problems, shortest path, spanning tree... and other problems for which
More informationMINI-TUTORIAL ON SEMI-ALGEBRAIC PROOF SYSTEMS. Albert Atserias Universitat Politècnica de Catalunya Barcelona
MINI-TUTORIAL ON SEMI-ALGEBRAIC PROOF SYSTEMS Albert Atserias Universitat Politècnica de Catalunya Barcelona Part I CONVEX POLYTOPES Convex polytopes as linear inequalities Polytope: P = {x R n : Ax b}
More informationBoolean constraint satisfaction: complexity results for optimization problems with arbitrary weights
Theoretical Computer Science 244 (2000) 189 203 www.elsevier.com/locate/tcs Boolean constraint satisfaction: complexity results for optimization problems with arbitrary weights Peter Jonsson Department
More informationAn approximation algorithm for the minimum latency set cover problem
An approximation algorithm for the minimum latency set cover problem Refael Hassin 1 and Asaf Levin 2 1 Department of Statistics and Operations Research, Tel-Aviv University, Tel-Aviv, Israel. hassin@post.tau.ac.il
More informationThe three-dimensional matching problem in Kalmanson matrices
DOI 10.1007/s10878-011-9426-y The three-dimensional matching problem in Kalmanson matrices Sergey Polyakovskiy Frits C.R. Spieksma Gerhard J. Woeginger The Author(s) 2011. This article is published with
More informationDijkstra s algorithm and L-concave function maximization
Noname manuscript No. (will be inserted by the editor) Dijkstra s algorithm and L-concave function maximization Kazuo Murota Akiyoshi Shioura Received: date / Accepted: date Abstract Dijkstra s algorithm
More informationOn disconnected cuts and separators
On disconnected cuts and separators Takehiro Ito 1, Marcin Kamiński 2, Daniël Paulusma 3 and Dimitrios M. Thilikos 4 1 Graduate School of Information Sciences, Tohoku University, Aoba-yama 6-6-05, Sendai,
More informationA Faster Strongly Polynomial Time Algorithm for Submodular Function Minimization
A Faster Strongly Polynomial Time Algorithm for Submodular Function Minimization James B. Orlin Sloan School of Management, MIT Cambridge, MA 02139 jorlin@mit.edu Abstract. We consider the problem of minimizing
More informationShow that the following problems are NP-complete
Show that the following problems are NP-complete April 7, 2018 Below is a list of 30 exercises in which you are asked to prove that some problem is NP-complete. The goal is to better understand the theory
More informationClassical Complexity and Fixed-Parameter Tractability of Simultaneous Consecutive Ones Submatrix & Editing Problems
Classical Complexity and Fixed-Parameter Tractability of Simultaneous Consecutive Ones Submatrix & Editing Problems Rani M. R, Mohith Jagalmohanan, R. Subashini Binary matrices having simultaneous consecutive
More informationDiskrete Mathematik und Optimierung
Diskrete Mathematik und Optimierung Annabell Berger, Winfried Hochstättler: Minconvex graph factors of prescribed size and a simpler reduction to weighted f-factors Technical Report feu-dmo006.06 Contact:
More informationComputability and Complexity Theory: An Introduction
Computability and Complexity Theory: An Introduction meena@imsc.res.in http://www.imsc.res.in/ meena IMI-IISc, 20 July 2006 p. 1 Understanding Computation Kinds of questions we seek answers to: Is a given
More informationA NEW SEMIDEFINITE PROGRAMMING HIERARCHY FOR CYCLES IN BINARY MATROIDS AND CUTS IN GRAPHS
A NEW SEMIDEFINITE PROGRAMMING HIERARCHY FOR CYCLES IN BINARY MATROIDS AND CUTS IN GRAPHS JOÃO GOUVEIA, MONIQUE LAURENT, PABLO A. PARRILO, AND REKHA THOMAS Abstract. The theta bodies of a polynomial ideal
More informationA NEW PROPERTY AND A FASTER ALGORITHM FOR BASEBALL ELIMINATION
A NEW PROPERTY AND A FASTER ALGORITHM FOR BASEBALL ELIMINATION KEVIN D. WAYNE Abstract. In the baseball elimination problem, there is a league consisting of n teams. At some point during the season, team
More informationM-convex Function on Generalized Polymatroid
M-convex Function on Generalized Polymatroid Kazuo MUROTA and Akiyoshi SHIOURA April, 1997 Abstract The concept of M-convex function, introduced recently by Murota, is a quantitative generalization of
More informationEnumerating minimal connected dominating sets in graphs of bounded chordality,
Enumerating minimal connected dominating sets in graphs of bounded chordality, Petr A. Golovach a,, Pinar Heggernes a, Dieter Kratsch b a Department of Informatics, University of Bergen, N-5020 Bergen,
More informationReductions. Reduction. Linear Time Reduction: Examples. Linear Time Reductions
Reduction Reductions Problem X reduces to problem Y if given a subroutine for Y, can solve X. Cost of solving X = cost of solving Y + cost of reduction. May call subroutine for Y more than once. Ex: X
More informationList H-Coloring a Graph by Removing Few Vertices
List H-Coloring a Graph by Removing Few Vertices Rajesh Chitnis 1, László Egri 2, and Dániel Marx 2 1 Department of Computer Science, University of Maryland at College Park, USA, rchitnis@cs.umd.edu 2
More informationDiscrete Applied Mathematics
Discrete Applied Mathematics 159 (2011) 1345 1351 Contents lists available at ScienceDirect Discrete Applied Mathematics journal homepage: www.elsevier.com/locate/dam On disconnected cuts and separators
More informationDual Consistent Systems of Linear Inequalities and Cardinality Constrained Polytopes. Satoru FUJISHIGE and Jens MASSBERG.
RIMS-1734 Dual Consistent Systems of Linear Inequalities and Cardinality Constrained Polytopes By Satoru FUJISHIGE and Jens MASSBERG December 2011 RESEARCH INSTITUTE FOR MATHEMATICAL SCIENCES KYOTO UNIVERSITY,
More informationFractional and circular 1-defective colorings of outerplanar graphs
AUSTRALASIAN JOURNAL OF COMBINATORICS Volume 6() (05), Pages Fractional and circular -defective colorings of outerplanar graphs Zuzana Farkasová Roman Soták Institute of Mathematics Faculty of Science,
More informationNotes for Lecture 2. Statement of the PCP Theorem and Constraint Satisfaction
U.C. Berkeley Handout N2 CS294: PCP and Hardness of Approximation January 23, 2006 Professor Luca Trevisan Scribe: Luca Trevisan Notes for Lecture 2 These notes are based on my survey paper [5]. L.T. Statement
More informationInteger Linear Programs
Lecture 2: Review, Linear Programming Relaxations Today we will talk about expressing combinatorial problems as mathematical programs, specifically Integer Linear Programs (ILPs). We then see what happens
More informationFRACTIONAL PACKING OF T-JOINS. 1. Introduction
FRACTIONAL PACKING OF T-JOINS FRANCISCO BARAHONA Abstract Given a graph with nonnegative capacities on its edges, it is well known that the capacity of a minimum T -cut is equal to the value of a maximum
More informationMin-Max Message Passing and Local Consistency in Constraint Networks
Min-Max Message Passing and Local Consistency in Constraint Networks Hong Xu, T. K. Satish Kumar, and Sven Koenig University of Southern California, Los Angeles, CA 90089, USA hongx@usc.edu tkskwork@gmail.com
More informationThe minimum G c cut problem
The minimum G c cut problem Abstract In this paper we define and study the G c -cut problem. Given a complete undirected graph G = (V ; E) with V = n, edge weighted by w(v i, v j ) 0 and an undirected
More information1 date: April 2, 2015 file: murota1. (B-EXC) For x, y B and for u supp + (x y), there exists v supp (x y) such that x χ u +
1 date: April 2, 2015 file: murota1 L-CONVEX FUNCTIONS AND M- CONVEX FUNCTIONS In the field of nonlinear programming (in continuous variables) convex analysis [22, 23] plays a pivotal role both in theory
More informationBichain graphs: geometric model and universal graphs
Bichain graphs: geometric model and universal graphs Robert Brignall a,1, Vadim V. Lozin b,, Juraj Stacho b, a Department of Mathematics and Statistics, The Open University, Milton Keynes MK7 6AA, United
More informationIntroduction to Linear and Combinatorial Optimization (ADM I)
Introduction to Linear and Combinatorial Optimization (ADM I) Rolf Möhring based on the 20011/12 course by Martin Skutella TU Berlin WS 2013/14 1 General Remarks new flavor of ADM I introduce linear and
More informationDeciding Emptiness of the Gomory-Chvátal Closure is NP-Complete, Even for a Rational Polyhedron Containing No Integer Point
Deciding Emptiness of the Gomory-Chvátal Closure is NP-Complete, Even for a Rational Polyhedron Containing No Integer Point Gérard Cornuéjols 1 and Yanjun Li 2 1 Tepper School of Business, Carnegie Mellon
More information2 Notation and Preliminaries
On Asymmetric TSP: Transformation to Symmetric TSP and Performance Bound Ratnesh Kumar Haomin Li epartment of Electrical Engineering University of Kentucky Lexington, KY 40506-0046 Abstract We show that
More informationLimitations of Algorithm Power
Limitations of Algorithm Power Objectives We now move into the third and final major theme for this course. 1. Tools for analyzing algorithms. 2. Design strategies for designing algorithms. 3. Identifying
More informationCS675: Convex and Combinatorial Optimization Fall 2014 Combinatorial Problems as Linear Programs. Instructor: Shaddin Dughmi
CS675: Convex and Combinatorial Optimization Fall 2014 Combinatorial Problems as Linear Programs Instructor: Shaddin Dughmi Outline 1 Introduction 2 Shortest Path 3 Algorithms for Single-Source Shortest
More informationTHE COMPLEXITY OF DISSOCIATION SET PROBLEMS IN GRAPHS. 1. Introduction
THE COMPLEXITY OF DISSOCIATION SET PROBLEMS IN GRAPHS YURY ORLOVICH, ALEXANDRE DOLGUI, GERD FINKE, VALERY GORDON, FRANK WERNER Abstract. A subset of vertices in a graph is called a dissociation set if
More informationMatroid Optimisation Problems with Nested Non-linear Monomials in the Objective Function
atroid Optimisation Problems with Nested Non-linear onomials in the Objective Function Anja Fischer Frank Fischer S. Thomas ccormick 14th arch 2016 Abstract Recently, Buchheim and Klein [4] suggested to
More informationDiscrete Convex Analysis
Hausdorff Institute of Mathematics, Summer School (September 1 5, 015) Discrete Convex Analysis Kazuo Murota 1 Introduction Discrete convex analysis [18, 40, 43, 47] aims to establish a general theoretical
More informationON COST MATRICES WITH TWO AND THREE DISTINCT VALUES OF HAMILTONIAN PATHS AND CYCLES
ON COST MATRICES WITH TWO AND THREE DISTINCT VALUES OF HAMILTONIAN PATHS AND CYCLES SANTOSH N. KABADI AND ABRAHAM P. PUNNEN Abstract. Polynomially testable characterization of cost matrices associated
More informationApproximability and Parameterized Complexity of Consecutive Ones Submatrix Problems
Proc. 4th TAMC, 27 Approximability and Parameterized Complexity of Consecutive Ones Submatrix Problems Michael Dom, Jiong Guo, and Rolf Niedermeier Institut für Informatik, Friedrich-Schiller-Universität
More informationOn the complexity of approximate multivariate integration
On the complexity of approximate multivariate integration Ioannis Koutis Computer Science Department Carnegie Mellon University Pittsburgh, PA 15213 USA ioannis.koutis@cs.cmu.edu January 11, 2005 Abstract
More informationRESEARCH ARTICLE. An extension of the polytope of doubly stochastic matrices
Linear and Multilinear Algebra Vol. 00, No. 00, Month 200x, 1 15 RESEARCH ARTICLE An extension of the polytope of doubly stochastic matrices Richard A. Brualdi a and Geir Dahl b a Department of Mathematics,
More informationDiscrepancy Theory in Approximation Algorithms
Discrepancy Theory in Approximation Algorithms Rajat Sen, Soumya Basu May 8, 2015 1 Introduction In this report we would like to motivate the use of discrepancy theory in algorithms. Discrepancy theory
More informationApproximating Submodular Functions. Nick Harvey University of British Columbia
Approximating Submodular Functions Nick Harvey University of British Columbia Approximating Submodular Functions Part 1 Nick Harvey University of British Columbia Department of Computer Science July 11th,
More informationMatroid Representation of Clique Complexes
Matroid Representation of Clique Complexes Kenji Kashiwabara 1, Yoshio Okamoto 2, and Takeaki Uno 3 1 Department of Systems Science, Graduate School of Arts and Sciences, The University of Tokyo, 3 8 1,
More informationLecture 21 (Oct. 24): Max Cut SDP Gap and Max 2-SAT
CMPUT 67: Approximation Algorithms Fall 014 Lecture 1 Oct. 4): Max Cut SDP Gap and Max -SAT Lecturer: Zachary Friggstad Scribe: Chris Martin 1.1 Near-Tight Analysis of the Max Cut SDP Recall the Max Cut
More informationAdvanced Combinatorial Optimization Updated April 29, Lecture 20. Lecturer: Michel X. Goemans Scribe: Claudio Telha (Nov 17, 2009)
18.438 Advanced Combinatorial Optimization Updated April 29, 2012 Lecture 20 Lecturer: Michel X. Goemans Scribe: Claudio Telha (Nov 17, 2009) Given a finite set V with n elements, a function f : 2 V Z
More informationGeneral Methods for Algorithm Design
General Methods for Algorithm Design 1. Dynamic Programming Multiplication of matrices Elements of the dynamic programming Optimal triangulation of polygons Longest common subsequence 2. Greedy Methods
More informationDiscrete Optimization 2010 Lecture 2 Matroids & Shortest Paths
Matroids Shortest Paths Discrete Optimization 2010 Lecture 2 Matroids & Shortest Paths Marc Uetz University of Twente m.uetz@utwente.nl Lecture 2: sheet 1 / 25 Marc Uetz Discrete Optimization Matroids
More information15-780: LinearProgramming
15-780: LinearProgramming J. Zico Kolter February 1-3, 2016 1 Outline Introduction Some linear algebra review Linear programming Simplex algorithm Duality and dual simplex 2 Outline Introduction Some linear
More informationCompetitive Equilibrium and Trading Networks: A Network Flow Approach
IHS Economics Series Working Paper 323 August 2016 Competitive Equilibrium and Trading Networks: A Network Flow Approach Ozan Candogan Markos Epitropou Rakesh V. Vohra Impressum Author(s): Ozan Candogan,
More informationSeparation Techniques for Constrained Nonlinear 0 1 Programming
Separation Techniques for Constrained Nonlinear 0 1 Programming Christoph Buchheim Computer Science Department, University of Cologne and DEIS, University of Bologna MIP 2008, Columbia University, New
More informationOn the adjacency matrix of a block graph
On the adjacency matrix of a block graph R. B. Bapat Stat-Math Unit Indian Statistical Institute, Delhi 7-SJSS Marg, New Delhi 110 016, India. email: rbb@isid.ac.in Souvik Roy Economics and Planning Unit
More informationDominator Colorings and Safe Clique Partitions
Dominator Colorings and Safe Clique Partitions Ralucca Gera, Craig Rasmussen Naval Postgraduate School Monterey, CA 994, USA {rgera,ras}@npsedu and Steve Horton United States Military Academy West Point,
More informationPreliminaries and Complexity Theory
Preliminaries and Complexity Theory Oleksandr Romanko CAS 746 - Advanced Topics in Combinatorial Optimization McMaster University, January 16, 2006 Introduction Book structure: 2 Part I Linear Algebra
More informationDisjunctive Temporal Reasoning in Partially Ordered Models of Time
From: AAAI-00 Proceedings Copyright 2000, AAAI (wwwaaaiorg) All rights reserved Disjunctive Temporal Reasoning in Partially Ordered Models of Time Mathias Broxvall and Peter Jonsson Department of Computer
More informationApproximation Algorithms for Maximum. Coverage and Max Cut with Given Sizes of. Parts? A. A. Ageev and M. I. Sviridenko
Approximation Algorithms for Maximum Coverage and Max Cut with Given Sizes of Parts? A. A. Ageev and M. I. Sviridenko Sobolev Institute of Mathematics pr. Koptyuga 4, 630090, Novosibirsk, Russia fageev,svirg@math.nsc.ru
More informationTORIC WEAK FANO VARIETIES ASSOCIATED TO BUILDING SETS
TORIC WEAK FANO VARIETIES ASSOCIATED TO BUILDING SETS YUSUKE SUYAMA Abstract. We give a necessary and sufficient condition for the nonsingular projective toric variety associated to a building set to be
More informationGraphical Model Inference with Perfect Graphs
Graphical Model Inference with Perfect Graphs Tony Jebara Columbia University July 25, 2013 joint work with Adrian Weller Graphical models and Markov random fields We depict a graphical model G as a bipartite
More informationSubmodular Functions and Their Applications
Submodular Functions and Their Applications Jan Vondrák IBM Almaden Research Center San Jose, CA SIAM Discrete Math conference, Minneapolis, MN June 204 Jan Vondrák (IBM Almaden) Submodular Functions and
More informationACO Comprehensive Exam March 20 and 21, Computability, Complexity and Algorithms
1. Computability, Complexity and Algorithms Part a: You are given a graph G = (V,E) with edge weights w(e) > 0 for e E. You are also given a minimum cost spanning tree (MST) T. For one particular edge
More informationExact Algorithms for Dominating Induced Matching Based on Graph Partition
Exact Algorithms for Dominating Induced Matching Based on Graph Partition Mingyu Xiao School of Computer Science and Engineering University of Electronic Science and Technology of China Chengdu 611731,
More informationMore on NP and Reductions
Indian Institute of Information Technology Design and Manufacturing, Kancheepuram Chennai 600 127, India An Autonomous Institute under MHRD, Govt of India http://www.iiitdm.ac.in COM 501 Advanced Data
More informationInterleaved Alldifferent Constraints: CSP vs. SAT Approaches
Interleaved Alldifferent Constraints: CSP vs. SAT Approaches Frédéric Lardeux 3, Eric Monfroy 1,2, and Frédéric Saubion 3 1 Universidad Técnica Federico Santa María, Valparaíso, Chile 2 LINA, Université
More informationShortest paths with negative lengths
Chapter 8 Shortest paths with negative lengths In this chapter we give a linear-space, nearly linear-time algorithm that, given a directed planar graph G with real positive and negative lengths, but no
More information